You’re seeing this ad based on the product’s relevance to your search query. If you are a seller and want to participate in this program click here to learn more.
Any questions? Our AI mode will help you find out quickly.
Try me now
Zaurion Duo Aqua · Dual-GPU Flagship
×2
Two Flagships, Pooled.
ABS Zaurion Duo Aqua Tower · ZAW9-3575X-RP6000MQ
Two NVIDIA RTX PRO 6000 MaxQ Blackwell cards — 96 GB of ECC GDDR7 each,
192 GB combined — on Intel's flagship Xeon W9-3575X server-board platform.
The models no single card here can hold, the multi-model fleets, the whole-team GPU shares:
this is the box they were waiting for.
192 GB CombinedDual 10 GbE + Management LAN2000 W 80 PLUS GoldAssembled in USA · TAA
The Headline Math
96 + 96, Honestly Counted
GPU 1
96 GB
RTX PRO 6000 MaxQ Blackwell
GDDR7 with ECC — a full flagship-class card with its own memory, Tensor Cores, and MIG partitions.
GPU 2
96 GB
RTX PRO 6000 MaxQ Blackwell
Identical twin — run it pooled with GPU 1, or dedicate it to its own model, team, or tenant.
=192 GBcombinedPooled via tensor/pipeline parallelism for one giant model — or split cleanly: one card per job. Per-card figures as listed; combined is simple arithmetic.
AI Silicon Spotlight · Hardware First
What Twin Flagships Change
GPUs · 2× RTX PRO 6000 MaxQ Blackwell
Capacity, Isolation, and Density — Pick All Three
192 GB combined — split the largest open-weight models across both cards, or give long-context serving room no single card here offers.
Perfect isolation — one card per production model means one model's traffic spike never starves the other.
Two cards of Universal MIG — partition both flagships into isolated instances; a whole team shares safely.
MaxQ density — the power-tuned editions are what let two flagship-class cards share one 2000 W 80 PLUS Gold supply, with headroom.
ECC GDDR7 on both cards — weights on always-on services get bit-flip protection, twice over.
Platform · Intel Xeon W9-3575X
A Server Backbone Built to Feed Two
Xeon W flagship silicon on the server-class Gigabyte MW83-RP0 board — the PCIe muscle two full-length flagships demand.
128 GB DDR5 ECC Registered — multi-day training and preprocessing pipelines finish without silent corruption.
2× 10 Gb/s LAN — two cards' worth of serving traffic, moved at wire speed; management LAN keeps IT out-of-band.
2 TB M.2 NVMe + 3.84 TB SSD — toolchains stay instant; datasets and checkpoints get their own lane.
W773-W80 tower airflow — engineered to keep twin GPUs at full song around the clock.
In Service
Three Jobs Only a Duo Does
01 · One Giant Model
Beyond Any Single Card Here
Tensor and pipeline parallelism spread one massive model across both cards — 192 GB combined holds what this family's single-GPU towers cannot, with context to spare.
Largest open-weight models · long context · private
02 · Two Production Models
A Fleet of Two, Fully Isolated
Chat model on GPU 1, embedding or vision model on GPU 2 — each with a full 96 GB to itself, each isolated from the other's load, both served over dual 10 Gb/s LAN.
One card per model · zero contention
03 · The Whole Team
Twin Cards, Many Tenants
Universal MIG on both cards partitions the duo into isolated instances for researchers, services, and experiments — a departmental GPU utility in one tower, managed out-of-band by IT.
2 × MIG partitions · management LAN
Scene photos are AI-generated illustrations of typical deployments, not photographs of this exact product.
Build to Order
Options Worthy of the Hardware
Optional
Pascari aiDAPTIV SSD
VRAM-expansion tier that offloads to flash — push large-model fine-tuning beyond even this duo's combined memory.
Optional
OpenClaw or NemoClaw
Enterprise AI agent framework pre-installed — agentic automation running the day the tower arrives.
Optional
Local LLM Package
Qwen 3.5 27B or higher, configured for private on-prem inference out of the box.
Full Config
As Configured
ZAW9-3575X-RP6000MQ · Zaurion Duo Aqua
GPUs
2× NVIDIA RTX PRO 6000 MaxQ Blackwell — 96 GB GDDR7 with ECC per card, 192 GB combined
CPU
Intel Xeon W9-3575X — workstation-class Xeon W platform, Socket on Gigabyte MW83-RP0
Each card carries its own 96 GB of GDDR7 with ECC. Modern stacks split one large model across both (tensor/pipeline parallelism), pool capacity for longer context, or dedicate one card per model.
Why MaxQ editions in a dual-GPU build?
Density and thermals: two power-tuned cards run side by side on a single 2000 W 80 PLUS Gold supply, with headroom, in a tower that lives outside a server room. Same 96 GB per card as the full-power edition.
Can it serve a whole team?
That's its natural habitat: two cards of Universal MIG partitions, dual 10 Gb/s LAN for wire-speed serving, and a management port so IT runs it like datacenter gear.
What software options can it ship with?
The Pascari aiDAPTIV SSD tier for VRAM expansion and fine-tuning, pre-installed OpenClaw or NemoClaw agent frameworks, and a Qwen 3.5 27B or higher local LLM package.
Government / public-sector eligible?
TAA compliant and assembled in the USA, with compliance documents available for purchase review.
How is it supported?
1-year parts and labor limited warranty backed by ABS manufacturer support. Build to Order — custom built and stress-tested after purchase.
192 GB combined GPU memory
Assembled in USA
TAA Compliant
Server-grade ECC platform
Per-card memory (96 GB GDDR7 with ECC) as listed; "192 GB combined" is the arithmetic total across both installed cards — usable capacity depends on the serving/parallelism stack.
ABS Zaurion Duo Aqua ZAW9-3575X-RP6000MQ · Gigabyte MW83-RP0 / W773-W80 platform · Specifications per the Newegg listing.
Optional components and software are configured at time of purchase.
Learn more about the ABS ZAW9-3575X-RP6000MQ
Model
Brand
ABS
Series
Zaurion Duo Aqua
Model
ZAW9-3575X-RP6000MQ
Part Number
ZAW9-3575X-RP6000MQ
Details
Workload
Workstation
Type
Tower
AI Features
AI Development: Accelerate AI development and inference workloads, and build agentic AI applications. Data Science: Accelerate your data pipelines. HPC: Accelerate scientific breakthroughs with enhanced accuracy. AI-Driven Rendering and Graphics: Deliver stunningly realistic designs and simulations with speed and precision. Video Content and Streaming: Boost performance for live media and video pipelines. Game Development: Drive new possibilities for game development with AI.
